Reduce the resource consumption of a complex image processing chain

From LRDE

Abstract

Even if the accuracy of the text localization system developed by the laboratory has strongly increased, one main problem remains the speed. Indeed, the CPU and the memory resources used by the system need to be reduced if we want to be able to use it to extract text from video or embedded it in resource limited devices like smartphones. To achieve this goal, various methods to reduce the memory and CPU consumption can be used. However all of these methods do not give the same results according to the input image and some of them can decrease the accuracy of the final result. The target is to present these methods, their results, and analyze in what kind of situation they can be used.